NVIDIA Quadro RTX 8000 GPU Review

Discussion in 'STH Main Site Posts' started by William Harmon, Nov 26, 2019.

  1. #1
  2. alex_stief

    alex_stief Active Member

    Joined:
    May 31, 2016
    Messages:
    446
    Likes Received:
    120
    Maybe it's just me, but the closing statement about the "psychological" reason to buy the highest end product possible sounds weird to me.
    It brings back memories of the "just buy it" article about RTX cards published on a different site.
    Call me stingy, but I would rather base hardware purchases on actual requirements. Not every CAD engineer needs the performance and memory size of this top-end GPU. I work with CAD from time to time, and I would shake my head if my company decided to go all-in on expensive hardware, in an attempt to motivate me.
    Don't get me wrong, I see the appeal of buying powerful hardware for qualified workers. But more than the price, I like being asked about my actual requirements.
     
    #2
    Last edited: Nov 26, 2019
  3. Cixelyn

    Cixelyn New Member

    Joined:
    Nov 7, 2018
    Messages:
    11
    Likes Received:
    9
    Not sure about CAD, but with deep learning workstations, I think the above "psychological" reason definitely applies.

    Model sizes have been constantly getting larger and larger for the past few yeras.
    It used to be that random pytorch & tensorflow repos would work with 4GB, now the minimum seems to be 8 or even 16GB for a single card. Today's perceived requirements might be obsolete within the span of a few months.

    For a research workstation, I'd definitely err on the side of adding more VRAM, as the workarounds for not having enough are rather complex.
     
    #3
  4. alex_stief

    alex_stief Active Member

    Joined:
    May 31, 2016
    Messages:
    446
    Likes Received:
    120
    Those are real requirements. Not buying expensive for the sake of expensiveness, as the article suggests.
     
    #4
  5. Patrick

    Patrick Administrator
    Staff Member

    Joined:
    Dec 21, 2010
    Messages:
    11,612
    Likes Received:
    4,566
    Just buy it is famous because no testing went into the statement. It was a blind recommendation on a site that used to be famous for data driven recommendations. That is why it is a famous line and very different than pointing out a psychological reason people purchase these cards after doing a review.

    I can tell you, I have spoken to a dozen local companies in the valley that will not buy their DL/ AI engineers less than Titan RTX's for workstations. There is such a big shortage for top talent, that 0-2 years out of school there are folks, that are great at computer vision as an example, making $400K+ in addition to equity-based comp. If those folks do not like the job, they just leave and go to another company that will pay them the same or more.

    In that context, $3000 or so per GPU to minimize frustrations and keep one person from jumping to a competitor or another company is huge. I have heard it now on numerous occasions.

    To me, the funnier part is that most of those same companies have large clusters. It is still effectively an HR move to keep people and attract others. These guys go to bars and tell everyone their company buys them Titan RTX/ Tesla V100/ DGX Stations/ Quadro RTX and those who have RTX 2080's become interested in joining a place with better hardware. If you have a 48GB card, and can run models in-memory that others at the bar cannot, people think your company is helping the data science. Keeping one person retained/ or generating a new hire lead pays for everyone's GPUs.
     
    #5
    nnunn, LukeP and T_Minus like this.
  6. William

    William Active Member

    Joined:
    May 7, 2015
    Messages:
    762
    Likes Received:
    236
    I think STH needs Titan RTX/RTX8000's in all reviewer workstations... yup that sounds good LOL :)
     
    #6
    Patrick likes this.
  7. AndrewH

    AndrewH New Member

    Joined:
    May 9, 2019
    Messages:
    2
    Likes Received:
    1
    Would you be able to provide a higher resolution version of the third image in the review (the back of the card's PCB)? I'm interested in what memory chips are being used and it's difficult to make out the text on them in the photo on the article.
     
    #7
  8. William

    William Active Member

    Joined:
    May 7, 2015
    Messages:
    762
    Likes Received:
    236
    #8
    AndrewH likes this.
  9. AndrewH

    AndrewH New Member

    Joined:
    May 9, 2019
    Messages:
    2
    Likes Received:
    1
    That's perfect thanks, if anyone else was curious they're Samsung K4ZAF325BM-HC14. 2GiB modules at 14Gbps running in 2 modules per "channel".
     
    #9
    William likes this.
  10. gigatexal

    gigatexal I'm here to learn

    Joined:
    Nov 25, 2012
    Messages:
    2,699
    Likes Received:
    500
    wait wait wait you can make almost half a mil doing computer vision? I am in the wrong field for sure
     
    #10
    Patrick likes this.
  11. Patrick

    Patrick Administrator
    Staff Member

    Joined:
    Dec 21, 2010
    Messages:
    11,612
    Likes Received:
    4,566
    @gigatexal Yes. Top tier talent in it is crazy now, especially with all of the self-driving car companies.
     
    #11
    gigatexal likes this.
  12. gigatexal

    gigatexal I'm here to learn

    Joined:
    Nov 25, 2012
    Messages:
    2,699
    Likes Received:
    500
    That’s awesome but before I could get up to speed in such a field the hype would likely have died down. Besides I enjoy databases too much
     
    #12
Similar Threads: NVIDIA Quadro
Forum Title Date
STH Main Site Posts NVIDIA Quadro P620 Low Profile, Low Power Graphics Card Review Sep 12, 2019
STH Main Site Posts New NVIDIA Quadro vDWS Up to 4 GPU VMs and vGPU Live Migration Oct 9, 2018
STH Main Site Posts NVIDIA Turing Introduced with the Quadro RTX Line Aug 14, 2018
STH Main Site Posts NVIDIA Quadro GV100 Workstation Monster Launched Mar 27, 2018
STH Main Site Posts New NVIDIA Tesla P6 and Quadro Virtual Data Center Workstation Software Aug 17, 2017

Share This Page